event: allow EPOLLET as event flag

[elogind.git] / src / libsystemd-bus / sd-event.c

diff --git a/src/libsystemd-bus/sd-event.c b/src/libsystemd-bus/sd-event.c
index ec18af43d51f385d90ccef2bcaca8cf58bb48c02..b5ddf71ebede89439e76adf1a3ba33fab830be05 100644 (file)
--- a/src/libsystemd-bus/sd-event.c
+++ b/src/libsystemd-bus/sd-event.c
@@ -584,7 +584,7 @@ _public_ int sd_event_add_io(
 
         assert_return(e, -EINVAL);
         assert_return(fd >= 0, -EINVAL);
 
         assert_return(e, -EINVAL);
         assert_return(fd >= 0, -EINVAL);

-        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP)), -EINVAL);
+        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL);

         assert_return(callback, -EINVAL);
         assert_return(ret, -EINVAL);
         assert_return(e->state != SD_EVENT_FINISHED, -ESTALE);
         assert_return(callback, -EINVAL);
         assert_return(ret, -EINVAL);
         assert_return(e->state != SD_EVENT_FINISHED, -ESTALE);


@@ -640,15 +640,16 @@ static int event_setup_timer_fd(
         }
 
         /* When we sleep for longer, we try to realign the wakeup to
         }
 
         /* When we sleep for longer, we try to realign the wakeup to

-           the same time wihtin each second, so that events all across
-           the system can be coalesced into a single CPU
-           wakeup. However, let's take some system-specific randomness
-           for this value, so that in a network of systems with synced
-           clocks timer events are distributed a bit. Here, we
-           calculate a perturbation usec offset from the boot ID. */
+           the same time wihtin each minute/second/250ms, so that
+           events all across the system can be coalesced into a single
+           CPU wakeup. However, let's take some system-specific
+           randomness for this value, so that in a network of systems
+           with synced clocks timer events are distributed a
+           bit. Here, we calculate a perturbation usec offset from the
+           boot ID. */

 
         if (sd_id128_get_boot(&bootid) >= 0)
 
         if (sd_id128_get_boot(&bootid) >= 0)

-                e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_SEC;
+                e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE;

 
         *timer_fd = fd;
         return 0;
 
         *timer_fd = fd;
         return 0;


@@ -983,7 +984,7 @@ _public_ sd_event_source* sd_event_source_unref(sd_event_source *s) {
         return NULL;
 }
 
         return NULL;
 }
 

-_public_ sd_event *sd_event_get(sd_event_source *s) {
+_public_ sd_event *sd_event_source_get_event(sd_event_source *s) {

         assert_return(s, NULL);
 
         return s->event;
         assert_return(s, NULL);
 
         return s->event;


@@ -1021,7 +1022,7 @@ _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events)
 
         assert_return(s, -EINVAL);
         assert_return(s->type == SOURCE_IO, -EDOM);
 
         assert_return(s, -EINVAL);
         assert_return(s->type == SOURCE_IO, -EDOM);

-        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP)), -EINVAL);
+        assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL);

         assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE);
         assert_return(!event_pid_changed(s->event), -ECHILD);
 
         assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE);
         assert_return(!event_pid_changed(s->event), -ECHILD);
 


@@ -1363,13 +1364,24 @@ static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) {
              dispatch as much as possible on the entire system.
 
           We implement this by waking up everywhere at the same time
              dispatch as much as possible on the entire system.
 
           We implement this by waking up everywhere at the same time

-          within any given second if we can, synchronised via the
+          within any given minute if we can, synchronised via the

           perturbation value determined from the boot ID. If we can't,
           perturbation value determined from the boot ID. If we can't,

-          then we try to find the same spot in every a 250ms
+          then we try to find the same spot in every 1s and then 250ms

           step. Otherwise, we pick the last possible time to wake up.
         */
 
           step. Otherwise, we pick the last possible time to wake up.
         */
 

-        c = (b / USEC_PER_SEC) * USEC_PER_SEC + e->perturb;
+        c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb;
+        if (c >= b) {
+                if (_unlikely_(c < USEC_PER_MINUTE))
+                        return b;
+
+                c -= USEC_PER_MINUTE;
+        }
+
+        if (c >= a)
+                return c;
+
+        c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC);

         if (c >= b) {
                 if (_unlikely_(c < USEC_PER_SEC))
                         return b;
         if (c >= b) {
                 if (_unlikely_(c < USEC_PER_SEC))
                         return b;


@@ -1412,6 +1424,9 @@ static int event_arm_timer(
         a = prioq_peek(earliest);
         if (!a || a->enabled == SD_EVENT_OFF) {
 
         a = prioq_peek(earliest);
         if (!a || a->enabled == SD_EVENT_OFF) {
 

+                if (timer_fd < 0)
+                        return 0;
+

                 if (*next == (usec_t) -1)
                         return 0;
 
                 if (*next == (usec_t) -1)
                         return 0;
 


@@ -1753,18 +1768,16 @@ _public_ int sd_event_run(sd_event *e, uint64_t timeout) {
         if (r < 0)
                 goto finish;
 
         if (r < 0)
                 goto finish;
 

-        if (event_next_pending(e) || e->need_process_child)
-                timeout = 0;
+        r = event_arm_timer(e, e->monotonic_fd, e->monotonic_earliest, e->monotonic_latest, &e->monotonic_next);
+        if (r < 0)
+                goto finish;

 
 

-        if (timeout > 0) {
-                r = event_arm_timer(e, e->monotonic_fd, e->monotonic_earliest, e->monotonic_latest, &e->monotonic_next);
-                if (r < 0)
-                        goto finish;
+        r = event_arm_timer(e, e->realtime_fd, e->realtime_earliest, e->realtime_latest, &e->realtime_next);
+        if (r < 0)
+                goto finish;

 
 

-                r = event_arm_timer(e, e->realtime_fd, e->realtime_earliest, e->realtime_latest, &e->realtime_next);
-                if (r < 0)
-                        goto finish;
-        }
+        if (event_next_pending(e) || e->need_process_child)
+                timeout = 0;

 
         m = epoll_wait(e->epoll_fd, ev_queue, EPOLL_QUEUE_MAX,
                        timeout == (uint64_t) -1 ? -1 : (int) ((timeout + USEC_PER_MSEC - 1) / USEC_PER_MSEC));
 
         m = epoll_wait(e->epoll_fd, ev_queue, EPOLL_QUEUE_MAX,
                        timeout == (uint64_t) -1 ? -1 : (int) ((timeout + USEC_PER_MSEC - 1) / USEC_PER_MSEC));